Face-Based Attention Recognition Model for Children with Autism Spectrum Disorder

Area of Science:

• Computer Science
• Developmental Psychology
• Biomedical Engineering

Background:

• Attention recognition is crucial for supporting children with autism spectrum disorders (ASD).
• Face-tracking offers unobtrusive automatic detection of attentional behaviors.
• ASD attentional complexity presents significant challenges for system development.

Purpose of the Study:

• To propose and evaluate a face-based attention recognition model for children.
• To compare geometric feature transformation with SVM against CNN approaches.
• To investigate model generalizability across participant types and task demands.

Main Methods:

• Developed a face-based attention recognition model using two distinct methods.
• Method 1: Geometric feature transformation with a Support Vector Machine (SVM) classifier.
• Method 2: Convolutional Neural Network (CNN) approach transforming time-domain spatial features to 2D images.

Main Results:

• The SVM classifier utilizing geometric feature transformation outperformed the CNN approach.
• Attention detection was more generalizable in typically developing children than in ASD groups.
• Attention detection was more generalizable in low-attention tasks compared to high-attention tasks.

Conclusions:

• Geometric feature transformation with SVM provides a robust method for attention recognition.
• Face-based attention recognition shows potential for real-time learning interventions in ASD.
• Further research can refine models for improved clinical attention interventions.